Difference between 32 and 64 bit on Advantech FWA-1305?
-
@cmb:
The problem is that from my experience, at least in a specific type of hardware (an Atom D410 firewall wth 6x 1Gbps ports and 2GB of DDR667 RAM), the system presents more throughput (from 500Mbps in 64bit to 650Mbps in 32bit) and much lower temperatures, both idle and under heavy load (a difference that varies between 20 to 10ºC, depending on the load).
Can someone give further feedback on these findings and if, having these observations in mind, if the advice to stay 64bit still applies?
That's the opposite of what you would expect to see. It's possible, though unlikely, that you have some weird combination of hardware where that is true for some reason. You'd be the first. I'd guess something else is at play, like a problem in testing methodology. Might want to start a new thread with your hardware specifics, what you're testing, and the specific results. It's been many years since I've seen a circumstance where running 32 bit on 64 bit hardware would be advisable. In recent years, the opposite is true - there are issues on 32 bit that don't exist on 64 bit, where running on 64 bit hardware.
Well, my "testing methodology", was simple empirical experience. Two similar systems (Advantech FWA-1305) with the specs described above that, with the same pfSense versions, performed better and with lower temperatures when running in 32-bit than when running in 64-bit, observed repeatedly in idle and under saturating bandwidth load. Any recommendations on an actual testing methodology?
-
Split this to its own thread as it's unrelated to the original.
How are you measuring throughput? Two different pieces of hardware could have different BIOS versions that impact things, maybe the fan on one of them isn't as good as the other, among other possibilities. Sounds like diff config files too, one maybe has more running than the other or one has powerd enabled and one doesn't.
-
They are both fanless systems with consecutive serial numbers, from the same manufacturing batch, with CF cards also from the same batch, same BIOS versions.
The routing throughput was being measured from pfSense's own graphing tool, as well as temperature.
Configuration files were the same. The problem is not between the two hardware systems. I mentioned both appliances, because I could replicate in both the same results. Both are faster and present lower temperatures (not only in the pfSense dashboard, you can also feel it to the touch) when running in 32 bit. I noticed this in one of the systems, when I tried using the 32 bit version instead of the 64 bit version that I was running before. Later, I was able to replicate this on the other machine.
-
What are you using to generate the traffic you're measuring is what I meant. If it's an unreliable/unpredictable method, like an Internet speed test site, that's not a reasonable means of comparing throughput of a device.
Do you have powerd enabled? System>Advanced, Misc. If so, what settings? If not, you probably want it enabled with hiadaptive.
-
Tested both with and without powerd enabled.
To generate traffic I was submitting it to iperf.
-
I have tested both again, in both combinations, now instead of using CF cards, I'm using equal SSDs, Kingston S200/30GB.
No changes in the results. Both devices present higher temperature and lower performance on 64 bit, than on 32 bit.